Overlaying two curves on mismatched axes can make unrelated signals look like they move together

What it is

When several variables are overlaid on one trend chart for comparison, inconsistent axes can make two unrelated curves look synchronized, or make two changes of very different magnitude look about the same size. In a dual-axis or multi-axis chart, the geometric alignment between curves is often just a coincidence produced by each axis's range being set independently — and that geometric alignment is not the same thing as the values actually being numerically comparable.

Why it happens

If each axis scales independently, the slope of any curve and the crossing point of any two curves can, in principle, be tuned to look however someone wants just by changing the axis limits — this is not an exaggeration; the same underlying data can be made to cross earlier or later, or look steeper or flatter, purely by rescaling. Meanwhile the visual system, scanning several curves at once, is naturally drawn to whether the curves move together before it registers each axis's own labeled scale. Once units, zero points, and direction are not handled consistently across the axes, the visual system delivers an intuitive verdict — "they move together," "they're proportional" — before any check of the axes happens, and in most cases that check never happens at all, so the wrong proportional or causal inference gets adopted straight from the visual impression.

Where it stops holding

This does not mean multi-axis charts should be banned outright: there are genuine cases with completely different units — temperature and pressure, say — where an operator needs to see whether they move together in time, and forcing both onto one shared axis would flatten one of the two signals into invisibility. So the real question is not whether multiple axes are allowed but whether they are constrained, clearly labeled, and offered alongside an alternative — converting both variables into a standardized deviation from their own baseline, for instance, or simply splitting them into two stacked small charts, time-aligned but with independent axes, which is often easier to read than forcing everything into one chart.

Applying it

Prefer shared units, standardized deviation, or time-aligned small multiples over dual-axis overlay wherever possible — all three support comparison without manufacturing geometric coincidence. When a dual axis is genuinely required, bind each curve's color firmly to its own axis, keep both axes increasing in the same direction, and always show the zero point and the full value range rather than a truncated segment. Testing should include a deliberately mismatched control chart — two genuinely unrelated curves whose axis ranges happen to align — to see whether viewers form a false co-movement judgment from it; if they do, the current axis design is itself producing the misreading.

Related

• Same group: Y3.07.1 Trend display for rate and direction · Y3.07.2 Adjustable trend time scale · Y3.07.4 Event-centered historical navigation
• Nearby: Y3.12 Low-saturation high-performance graphics · Y1.03 Trends and rate of change
• Search terms: dual-axis chart · axis scaling · misleading graph · small multiples